Arabic Language Sentiment Analysis Using Feature Engineering and Deep Learning RNN-LSTM Framework

Abstract

Consistent enormous generated data on the internet becomes one of the most sophisticated tasks that need profound artificial analysis. Given that, the researches in sentiment analysis addressed many techniques to deal with sentiment analysis (SA) in multiple languages. Nonetheless, the state-of-the-art for the Arabic Language sentiment analysis (ALSA) quite needs more improvements. The research area in Arabic has many challenges on account of its complex, unique nature and structure. SA is the computational analysis of the people’s opinions, attitudes, emotions and evaluations from the document. This paper proposes an approach for examining the SA in the Arabic language using the linguistic feature extraction, word embedding and deep learning RNN-LSTM frameworks on the sentence level. The proposed model has been evaluated on a large Dataset of Arabic Tweets (ArSAS) reaching 21 thousand Arabic tweets twice. The first experiment was without considering the linguistic features and compared to extract the linguistics features from the data. The experiment demonstrates that the approach achieves state-of-the-art results and it shows a significant increase in the F-score reaching 81.1%.